Skip to content
All library documents

Solving the Affine Short-Rate Riccati Equation

Article Quant Q&A · Author: lrh09

Summary

The document derives the bond-pricing coefficient for a short-rate process with linear drift and square-root-affine variance. It starts from the Riccati ordinary differential equation for the coefficient B, rescales B to obtain another Riccati equation, and applies a logarithmic-derivative substitution to convert that equation into a second-order linear equation with constant coefficients. Solving the characteristic equation gives exponential components; the terminal condition at maturity determines their relative weights. Substitution then recovers the closed-form expression for B as a function of time to maturity and the model parameters.

This derivation explains a reusable way to solve Riccati equations in affine term-structure models. It also provides the resulting formula and connects its discriminant notation to the expression in the question. The treatment assumes the stated model and terminal condition, and it focuses on B rather than deriving the full bond price or discussing parameter estimation, numerical stability, or empirical fit.

Key ideas

  • A rescaling preserves the Riccati structure while simplifying the coefficient equation.
  • A logarithmic-derivative substitution turns the Riccati equation into a linear second-order equation.
  • The characteristic roots yield exponential solutions that determine the closed form for B.
  • The maturity boundary condition fixes the constants in the general solution.

Tags

Full text
# Riccati Equation in spot rate model


# Riccati Equation in spot rate model












Given that $dr=(\eta-\gamma r)dt+\sqrt{\alpha r+\beta}dW$

Let $Z(r,t)=e^{A(t;T)-rB(t;T)}$,

\begin{matrix} \frac{dA}{dt}=\eta B-\frac{1}{2}\beta {{B}^{2}} \\ \frac{dB}{dt}=\frac{1}{2}\alpha {{B}^{2}}+\gamma B-1 \\ \end{matrix}

How can we solve for $B$ in this case?

$B$ is a Riccati Equation. The answer for B is given below:

$B(t;T)=\frac{2(e^{\Psi_1(T-t)}-1)}{(\gamma+\Psi_1)(e^{\Psi_1(T-t)}-1)+2\Psi_1}$,

where $\Psi_1=\sqrt{\gamma^2+2\alpha}$

## Answer by LocalVolatility (score 5, accepted)

https://quant.stackexchange.com/a/34758

As you noted, this is a Riccati type ODE and it can thus be simplified using the standard transformations for this class - see e.g. Wikipedia. We start by defining

\begin{equation} C(t, T) = \frac{1}{2} \alpha B(t, T) \qquad \Rightarrow \qquad C_t(t, T) = \frac{1}{2} \alpha B_t(t, t) \end{equation}

and get

\begin{eqnarray} C_t(t, T) & = & C^2(t, T) + \gamma C(t, T) - \frac{1}{2} \alpha. \end{eqnarray}

We now set

\begin{eqnarray} C(t, T) = -\frac{D_t(t, T)}{D(t, T)} \qquad \Rightarrow \qquad C_t(t, T) & = & -\frac{D_{tt}(t, T)}{D(t, T)} + \frac{D_t^2(t, T)}{D^2(t, T)}\\ & = & \frac{-D_{tt}(t, T)}{D(t, T)} + C^2(t, T) \end{eqnarray}

and get

\begin{eqnarray} \frac{D_{tt}(t, T)}{D(t, T)} & = & C^2(t, T) - C_t(t, T)\\ & = & -\gamma C(t, T) + \frac{1}{2} \alpha\\ & = & \gamma \frac{D_t(t, T)}{D(t, T)} + \frac{1}{2} \alpha \end{eqnarray}

or

\begin{equation} D_{tt}(t, T) = \gamma D_t(t, T) + \frac{1}{2} \alpha D(t, T). \end{equation}

This is a homogeneous second order linear ODE with constant coefficients and can be solved using standard methods. We note that $T$ has been fixed and make another substitution by defining $\tau = T - t$ such that $E(\tau) = D(t, T)$. We get

\begin{equation} E_{\tau \tau}(\tau) + \gamma E_\tau(\tau) - \frac{1}{2} \alpha E(\tau) = 0. \end{equation}

The characteristic equation is

\begin{equation} r^2 + \gamma r - \frac{1}{2} \alpha = 0 \end{equation}

with solutions

\begin{equation} r_{1, 2} = -\frac{1}{2} \gamma \pm\frac{1}{2} \sqrt{\gamma^2 + 2 \alpha} := \beta \pm \lambda. \end{equation}

Note that $\lambda = \frac{1}{2} \psi_1$ in your notation. We thus have the general solution

\begin{equation} E(\tau) = c_1 e^{(\beta + \lambda) \tau} + c_2 e^{(\beta - \lambda) \tau} \end{equation}

with

\begin{equation} E_\tau(\tau) = (\beta + \lambda) e^{(\beta + \lambda) \tau} + (\beta - \lambda) c_2 e^{(\beta - \lambda) \tau} \end{equation}

and for some constants $c_1$ and $c_2$ that have to be determined. We obtain the solution to the Riccati ODE by substituting back

\begin{equation} B(t, T) = \frac{2 C(t, T)}{\alpha} = -\frac{2 D_t(t, T)}{\alpha D(t, T)} = \frac{2 E_\tau(\tau)}{\alpha E(\tau)}. \end{equation}

Applying the terminal condition yields

\begin{eqnarray} B(T, T) = 0 \qquad & \Leftrightarrow & \qquad E_\tau(0) = 0\\ & \Leftrightarrow & \qquad c_1 = -c_2 \frac{\beta - \lambda}{\beta + \lambda}. \end{eqnarray}

Thus

\begin{equation} E(\tau) = \frac{c_2}{\beta + \lambda} e^{\beta \tau} \left( (\beta + \lambda) e^{-\lambda \tau} - (\beta - \lambda) e^{\lambda \tau} \right) \end{equation}

and

\begin{equation} E_\tau(\tau) = (\beta - \lambda) c_2 e^{\beta \tau} \left( e^{-\lambda \tau} - e^{\lambda \tau} \right). \end{equation}

We finally get

\begin{eqnarray} B(t, T) & = & \frac{2 (\beta^2 - \lambda^2) \left( e^{-\lambda \tau} - e^{\lambda \tau} \right)}{\alpha \left( (\beta + \lambda) e^{-\lambda \tau} - (\beta - \lambda) e^{\lambda \tau} \right)}\\ & = & \frac{\left( e^{2 \lambda \tau} - 1 \right)}{(\beta + \lambda) - (\beta - \lambda) e^{2 \lambda \tau}}\\ & = & \frac{2 \left( e^{\psi_1 (T - t)} - 1 \right)}{\left( \gamma + \psi_1 \right) \left( e^{\psi_1 (T - t)} - 1 \right) + 2 \psi_1}. \end{eqnarray}

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.